Distillation技術を用いたネットワークの分類精度に対する考察

URI http://harp.lib.hiroshima-u.ac.jp/pu-hiroshima/metadata/12595
File
Title
Distillation技術を用いたネットワークの分類精度に対する考察
Title Alternative
A Consideration of Classification Accuracy of Network Using Distillation Technology
Author
氏名 藤井 哲崇
ヨミ フジイ ヨシタカ
別名 Fujii Yoshitaka
氏名 市村 匠
ヨミ イチムラ タクミ
別名 Ichimura Takumi
Abstract

Abstract—Since the architecture of deep learning has many parameters such as weights, bias, and its structure, we may be faced with a problem that the learning technology can not be implemented in a real world. The general deep learning method takes a long time calculation. Therefore, a new technology for the simplification of architecture is required. The trained simplified network should reach the higher classification capability under the optimal structure. In this paper, we investigate the Distillation technology proposed by Hinton using AIC to evaluate the trained network architecture and its accuracy of classification.

Description

開催日:平成29年7月22日
会場:広島工業大学

Journal Title
2017 IEEE SMC Hiroshima Chapter若手研究会講演論文集
Spage
49
Epage
53
Published Date
2017
Publisher
IEEE SMC Hiroshima Chapter
Contributor
市村匠
Language
jpn
NIIType
Conference Paper
Text Version
出版社版
Set
pu-hiroshima